relay-knowledge 1.1.17

Graph-database-based knowledge graph project.
Documentation 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134

//! Deterministic local semantic, vector, and lexical scoring primitives.

use std::collections::BTreeSet;

pub(super) use crate::identity::stable_hash64;
use crate::retrieval::terms::extend_normalized_terms;

pub(super) fn token_signature(
    content: &str,
    labels: &[String],
    source_path: Option<&str>,
) -> Vec<String> {
    let mut terms = BTreeSet::new();
    collect_terms(content, &mut terms);
    collect_terms(&labels.join(" "), &mut terms);
    collect_terms(source_path.unwrap_or_default(), &mut terms);

    terms.into_iter().collect()
}

fn collect_terms(value: &str, terms: &mut BTreeSet<String>) {
    extend_normalized_terms(value, 2, terms);
}

pub(super) fn hashed_vector(
    content: &str,
    labels: &[String],
    source_path: Option<&str>,
    dimension: usize,
) -> Vec<f64> {
    if dimension == 0 {
        return Vec::new();
    }
    let terms = token_signature(content, labels, source_path);
    let mut vector = vec![0.0; dimension];
    for term in terms {
        let hash = stable_hash64(term.as_bytes());
        let index = (hash as usize) % dimension;
        let sign = if hash & 1 == 0 { 1.0 } else { -1.0 };
        vector[index] += sign;
    }
    normalize_vector(&mut vector);

    vector
}

fn normalize_vector(vector: &mut [f64]) {
    let norm = vector.iter().map(|value| value * value).sum::<f64>().sqrt();
    if norm == 0.0 {
        return;
    }
    for value in vector {
        *value /= norm;
    }
}

pub(super) fn semantic_overlap_score(
    query_terms: &BTreeSet<String>,
    document_terms: &BTreeSet<String>,
) -> f64 {
    if query_terms.is_empty() || document_terms.is_empty() {
        return 0.0;
    }
    let intersection = query_terms.intersection(document_terms).count();
    if intersection == 0 {
        return 0.0;
    }
    let union = query_terms.union(document_terms).count();

    intersection as f64 / query_terms.len() as f64 + intersection as f64 / union as f64
}

pub(super) fn cosine_similarity(left: &[f64], right: &[f64]) -> f64 {
    if left.len() != right.len() || left.is_empty() {
        return 0.0;
    }
    left.iter()
        .zip(right.iter())
        .map(|(left, right)| left * right)
        .sum::<f64>()
        .max(0.0)
}

pub(super) fn overlap_score(
    query: &str,
    content: &str,
    labels: &[String],
    source_path: Option<&str>,
) -> f64 {
    let haystack = format!(
        "{} {} {}",
        content.to_lowercase(),
        labels.join(" ").to_lowercase(),
        source_path.unwrap_or_default().to_lowercase()
    );
    let mut score = 0.0;
    for token in query.to_lowercase().split_whitespace() {
        if haystack.contains(token) {
            score += 1.0;
        }
    }
    if score > 0.0 {
        return score;
    }

    identifier_overlap_score(query, content, labels, source_path)
}

fn identifier_overlap_score(
    query: &str,
    content: &str,
    labels: &[String],
    source_path: Option<&str>,
) -> f64 {
    let query_terms = token_signature(query, &[], None);
    let document_terms = token_signature(content, labels, source_path);
    query_terms
        .iter()
        .filter(|term| {
            let term = term.as_str();
            document_terms
                .iter()
                .any(|candidate| candidate == term || fuzzy_identifier_part_match(term, candidate))
        })
        .count() as f64
}

fn fuzzy_identifier_part_match(query_term: &str, candidate: &str) -> bool {
    query_term.len() >= 3 && candidate.len() >= 3 && candidate.contains(query_term)
}

#[cfg(test)]
#[path = "mod_tests.rs"]
mod tests;